Lawyers · chartered accountants · notaries · family offices
A private AI, installed on your premises
Summarise a case file, query your archives, prepare a first draft of a letter — with a Mistral model running on your own server. No client data leaves the practice.
Why a local AI?
With an online AI assistant, every question and every document sent goes to an external provider. For a profession bound by professional secrecy, this is precisely the delicate point: the Conseil national des barreaux (French national bar council) guide on ethics and AI (2026) reminds us that any request sent to an online generative AI transmits data to the company that operates it.
An AI installed on a server that belongs to you, on your premises, removes this transfer:
- documents, questions and answers stay on your hardware;
- access is limited to your network, with per-user accounts;
- no per-user subscription, no query quota;
- you can unplug the Internet cable: the AI keeps working.
Read more: Online AI and professional secrecy — the real risks.
What the AI does for your team
Search in your documents
Ask a question in plain language: the AI searches your templates, your consultations, your notes and your internal procedures, and cites the documents it relies on.
Case-file summaries
Summary of a case file, chronology of events, list of documents, points to check — in a few minutes instead of an hour.
First drafts
Letters, emails, summary notes, templates for deeds or pleadings from your own models. The professional rereads, corrects and signs.
Translation and languages
Documents in English, German, Italian or Russian translated and summarised in-house — useful in Alsace, Monaco and the French Riviera.
The AI prepares, the professional decides. It can be wrong: it replaces neither the verification nor the responsibility of the person who signs.
The models: Mistral, locally
Mistral AI is a French company. Several of its models are published as open weights under the Apache 2.0 licence, which allows professional use on your own hardware, royalty-free.
| Model | Strengths | Suitable hardware |
|---|---|---|
| Mistral Small 3.2 (24 billion parameters) | Fast, very good in French, also reads images and scanned documents | Compact server with a 24 GB graphics card |
| Mistral Small 4 (119 billion parameters, "mixture of experts" architecture) | Deeper reasoning, very long context (256,000 tokens) | Machine with 128 GB of unified memory |
I choose the model with you according to your uses, the number of staff and your budget — and I show you a demonstration on your own types of documents before any decision.
The hardware
Three typical configurations. The hardware is bought directly by the practice from the supplier: it belongs to you, with its manufacturer warranty. I give you the exact configuration and install it.
| Configuration | For whom | Indicative hardware budget |
|---|---|---|
| Mini server (Minisforum MS-02 Ultra) + NVIDIA RTX PRO 4000 Blackwell 24 GB card | Practice of 1 to 5 people | around €4,000 to €5,000 |
| Mini PC AMD Ryzen AI Max+ 395, 128 GB | Practice wanting a larger model on a tight budget | around €2,000 to €2,500 |
| NVIDIA DGX Spark, 128 GB | Team of 3 to 10 simultaneous users | around €6,000 to €7,500 |
Prices observed in October 2026, subject to sharp change (memory shortage). Consumption of a few tens to a few hundred watts, on an ordinary socket, with no server room.
The offer
"Private AI at your office" installation — from €2,900 excl. VAT (hardware not included)
- Scoping workshop: uses, documents concerned, access rights
- Installation of the server, the Mistral model and a web interface for the team
- Indexing of your documents (templates, procedures, selected archives)
- Per-user accounts, disk encryption, backup, access log
- Team training (half a day) and user guide
- 30 days of adjustments
Maintenance — from €200 excl. VAT / month
Security and model updates, monitoring, verified backups, support. On-site intervention in Strasbourg, Alsace, Nice and Monaco.
Frequently asked questions
Is a local AI as good as ChatGPT?
For the everyday tasks of a practice — summarising, searching documents, writing a first draft in French — a recent Mistral model installed locally gives very good results. For very complex reasoning, the largest online models remain ahead. I show you the difference on your own examples before any order.
How many people can use it at the same time?
It depends on the model and the hardware: from a few people for a mini server to about ten for a 128 GB machine. I size it according to your number of staff.
What happens if the hardware fails?
The documents remain on your own backups; the hardware is under manufacturer warranty. With maintenance, I bring the installation back into service on replacement hardware.
What about the EU AI regulation and the GDPR?
A local installation greatly simplifies the question of data transfer. I document the processing (purposes, data, access, retention periods) for your GDPR register. Team training also meets the AI proficiency obligation under the EU regulation (article 4).
Which task would you like to stop doing by hand?
Describe it in two lines on WhatsApp or Telegram: I reply within one business day with a first approach and a price range.